How do I delete a TRAE CLI session?

Remove its rollout file under ~/.trae/cli/sessions/YYYY/MM/DD/; traex archive only moves one into archived_sessions/, it does not delete it. The sessions directory is dated, so one day's work is one directory, and rm -rf ~/.trae/cli/sessions removes every session. Prompt history is a separate file, ~/.trae/cli/history.jsonl, and survives that.

Where does TRAE CLI store its sessions?

Under ${TRAE_HOME:-~/.trae}/cli, where TRAECLI_HOME moves that directory and TRAE_HOME moves the home above it: rollouts at sessions/YYYY/MM/DD/rollout-*.jsonl as Codex-shaped JSONL, archived ones in archived_sessions/, and prompts in history.jsonl.

Does traex archive delete a session?

No. It moves the rollout into archived_sessions/ beside sessions/, and the prompts stay in history.jsonl; the bytes leave the disk only when you remove the files.

Does deleting a TRAE rollout remove the prompt from history?

No. history.jsonl holds your prompts independently of the rollouts, with no assistant replies, so a prompt can survive the session it belongs to. Delete both if you want it gone.

How to delete TRAE CLI sessions, and what you lose

they are rollout files under ~/.trae/cli/sessions, one per session, and prompt history is a separate file beside them

TRAE CLI writes each session to disk as it runs. 2.0 ships as traex and is a closed-source fork of codex-rs, so the files are Codex rollouts under TRAE's own home instead of ~/.codex. The store is ${TRAE_HOME:-~/.trae}/cli, and there are two separate things in it: the rollouts, which are the conversations, and history.jsonl, which is your prompts. Deleting one does not touch the other.

How to delete TRAE CLI sessions

traex resume <id> lists and reopens sessions, and traex archive <id> hides one without deleting it — the rollout moves to archived_sessions/ and stays on disk. Beyond that the sessions are files. The path is dated, so a day of work is a directory, and a rollout that has been compressed ends in .jsonl.zst, which a pattern ending in .jsonl does not match.

# what is there, newest first
ls -t ~/.trae/cli/sessions/*/*/*/rollout-*.jsonl* | head

# one session — the id is what traex resume takes
rm ~/.trae/cli/sessions/2026/10/17/rollout-*<id>.jsonl*

# one day
rm -rf ~/.trae/cli/sessions/2026/10/17

# every session, including the ones traex archived
rm -rf ~/.trae/cli/sessions ~/.trae/cli/archived_sessions

# your prompts, which the lines above do not touch
rm ~/.trae/cli/history.jsonl

If TRAECLI_HOME is set it moves the whole cli directory, and TRAE_HOME moves the home above it; all of the above moves with either.

What is in a rollout

JSON Lines, appended as the session runs. The first record is session_meta and carries the session id and the working directory; the rest are Codex's own events — turn_context, response_item, event_msg — plus TRAE's history_mutation records for tool calls, which Codex does not write. Because the file is appended to while the session is live, the last line of an interrupted session can be torn — a reader has to tolerate that, and rm does not care either way.

It is plain text and it is verbatim: whatever a command printed into the conversation is in the file. Treat the directory the way you treat your shell history. The format registry documents every field.

An archived session is not deleted

traex archive <id> moves one rollout into archived_sessions/ beside sessions/; the bytes are all still there, and so is the prompt in history.jsonl. Nothing leaves the disk until someone runs rm, which is why the commands above name the archive directory as well when they mean “everything”.

What you lose

The rollouts are what any later search over your own work reads — including traex resume, which lists what is on disk. Delete them and the record of what the agent ran goes with them: the commands, their output, and the edits, none of which are recoverable from a summary.

deja reads these files with the Codex reader and files them under trae, so a decision made in TRAE answers a question asked in Claude Code or Cursor. A session it has already indexed stays searchable after the file goes; only deja forget drops one on purpose. In the screen bare deja opens, a deleted session sits under the Deleted tab — but TRAE is one of the agents deja does not write a rollout back for, because TRAE CLI records turns its own way since 0.208 and no rebuilt rollout has been checked against it yet. That is a reason to index before deleting, not a reason to keep gigabytes.

The same question for Codex · Where every agent stores history · The same question for Claude Code · The TRAE CLI format registry

Found this useful? Star deja-vu on GitHub.